Predictive Experiment for Twitter sentiment analysis

May 22, 2016


Report Abuse
This experiment demonstrates the use of the Execute R Script, Feature Selection, Feature Hashing modules to train a text sentiment classification engine.
This specific experiment is a second step in creating a web service for [twitter sentiment analysis]( In this step we will add a column transform that was saved as part of the previous training experiment to make sure filter based feature selection doesn't changes the feature set used to train the original model The steps to accomplish this: 1. Go to your [training experiment]( and save the output of "Select Column Transform" as a new "FilterTransform" 2. Go to your [training experiment]( and save the output of "Trained Model" as a new "TrainedModel" 3. Save your training experiment as "Binary Classification: Twitter sentiment analysis [Predictive Exp.]" 4. Remove train, "select columns transform" , "Filter based feature", split, evaluate and one of the score module 5. Use project column to drop "sentiment label" from reader and Modify execute R script to remove "sentiment label" 6. Drag and drop previously saved transform from step 1. (All saved transforms can be accessed under "Transforms" menu on the left hand navigation pane) 7. Drag and drop apply transform module 8. Output of feature hashing to the left port of apply transform module and connect transform module from step 5 to right port of apply transform 9. Drag and drop previously saved trained model from step 2. (All saved models can be accessed under "Trained Models" menu on left) 10. Connect trained model to the left port of score model and connect output of "apply transform" module to the right port of score model 8. Drag and drop web service input and output. Connect web service input to the input of "Execute R" module and connect output of score module to web service output Once above steps are completed, your graph will look like the one shown below ![](